Intelligent steel bridge coating assessment using neuro-fuzzy recognition approach

Recently, digital image recognition has been applied to steel bridge coating assessment. However, non-uniform illumination is always a problem and affects the accuracy of processed results. In order to resolve the recognition problems arising from non-uniformly illuminated images, the neuro-fuzzy recognition approach (NFRA) is proposed. NFRA segments a grayscale image into three areas in accordance with the illumination values of the pixels in the image. The three average illumination values of the three areas are then sent to a pre-trained neural network to generate three corresponding threshold values. The fuzzy adjustment system adjusts the gray level values of the pixels along the boundaries between areas. Finally, image thresholding is applied to get a binary image containing only the object pixels and the background pixels. Besides the background and framework of NFRA, this paper also introduces a random sampling plan to help select positions for image taking during steel bridge coating inspections.